Closed RichardWaiteSTFC closed 4 weeks ago
:warning: Please install the to ensure uploads and comments are reliably processed by Codecov.
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 42.65%. Comparing base (
1b52407
) to head (479e28f
).
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
4 files ± 0 116 suites +4 4m 55s :stopwatch: +19s 731 tests +11 713 :white_check_mark: +11 18 :zzz: ±0 0 :x: ±0 2 096 runs +44 2 060 :white_check_mark: +44 36 :zzz: ±0 0 :x: ±0
Results for commit 479e28f8. ± Comparison against base commit 1b524075.
:recycle: This comment has been updated with latest results.
@mducle Thanks for the review, I have done most of the changes requested, in addition in https://github.com/SpinW/spinw/pull/198/commits/479e28f898bde4ced589a207ba617d9655049475 I now output parameter errors as a vector the same size as the input parameters (with 0 error for parameters which were fixed). I thought this was more standard/expected behaviour.
I think there are only 2 points still to be ironed out:
@RichardWaiteSTFC Thanks for this.
Let's keep the current auto step-size algorithm. If we find when we use it that it's too slow then we can come back and revisit this and maybe update it.
I'm happy to leave the in-place changing of the parameter vector.
Description
New function in ndbase to calculate the hessian matrix given a cost function and parameters. The hessian matrix evaluated at the minimum can be used to calculate the covariance matrix and hence the uncertainties on fit parameters (this is useful because not all ndbase optimizers return uncertainties).
To-Do
To Test
(1) Run this to compare the hessian to that calculated using the jacobian of the residuals from
lsqnonlin